|Philosophy of Mind
Qualia, Robots and Complementarity
ABSTRACT: Jackson claims that a person who sees colors for the first time by this very fact acquires a certain knowledge which she or he could not have learned in a black and white world. This argument can be generalized to other secondary qualities. I argue that this claim is indefensible without implicit recourse to the first-person experience; also Nagels "what it is like" argument is polemically weak. Hence, we have no argument able to dismiss physicalism by consideration of first-person qualia (contra Jackson); however, it does not force us to endorse qualia-reductionism. In the second part of my paper I defend non-reductionism in a different way. Following Nagel and Harman, I try to avoid criticisms usually presented against Nagel, seeing subjectivity and objectivity as two complementary structures of the subjective and objective element of our language. I refer to classical German philosophy, phenomenology and Marxist dialectics which have developed a complementary approach crucial in the reductionist/anti-reductionist controversy in the philosophy of mind.
"Opinion says hot and cold, but the reality is atoms and empty space."
Jacksons Black and White Mary (1) case illustrates an argument that our first-person experience of qualia provides knowledge not accessible through third-person means. This argument seems suspicious: if knowledge, by definition, needs to be grasped in third-person terms, it is inconceivable how an exclusively first-person experience may give us knowledge.
Harman (92) develops complementarity of subjective and objective aspects within his functionalism of concepts. This notion needs to be extended to epistemic complementarity. I refer to Classical German Philosophy, Phenomenology and Marxism which have developed a complementary approach crucial in the reductionist anti-reductionist controversy in philosophy of mind.
I. The Knowledge Argument
As Jackson (86) emphasizes in his polemics with Churchland (85) the main point of the case of Black and White Mary is not that Mary cannot imagine what it is like to sense red but that "she would not know" (2) how red things look. I accept this argument, but not the next step. Jackson argues: "But if physicalism is true she would know; and no great powers of imagination would be called for." (3)
There are two versions of physicalism: the first, against which Jackson has something to say, maintains that every instance of knowledge can be couched in third-person language. (4) But the assumption that physical information is solely what "you can tell" is not a necessary condition of physicalism. Jackson's criticism does not reach another version of physicalism which assumes that everything is physical but says nothing about it being describable. Physicalism of the second kind presupposes that "there is no difference without physical difference" (5) but it does not imply that every physical difference can be described in language.
That the knowledge argument works against the first but not against the second form of physicalism can be seen in Jeckson's reply to one of Churchlands (6) objections:
The conclusion (3)' follows from premisses (1)' and (2)' only if the first kind of physicalism is assumed; otherwise either premise (1) is unsatisfied, or conclusion (3) goes too far. Before we examine these possibilities it is worth pointing out that in most of the literature premise (2) is criticized; Mary is said to learn nothing. But (2) seems undeniable since she does learn a new capacity, which is also Nemirows and Lewis(90) point.
The First Criticism
Nemirow argues (8) that knowing 'what it is like' is not propositional knowledge but a linguistically inexpressible ability like wiggling ones ears or riding a bicycle; it is not knowledge. This objection to the knowledge argument is supposed to dismiss Jackson's Knowledge Argument. Obvioulsy, this objection does not hit Nagel's argument which does not take first-preson experience to be justifiable by third-person knowledge.
Nemirow's argument is tacitly based on a tertium non datur distinction between "knowledge that may be summarized propositionally" and "abilities" which are supposed not to be knowledge and therefore be somehow irrelevant for questions of epistemology, like qualia. As the second type of physicalism demonstrates this dichotomy is untenable. Knowledge how is a sort of knowledge (the only kind of intersubjectively verifiable knowledge that Mary did not have); whether it can be "summarized propositionally" is irrelevant for the present argument. Mary did not know "everything there is to know about other people" and consequently premise (1) is unsatisfied whereas premise (2) is satisfied.
We may use the notion of knowledge* in a sense in which whatever is know* is sayable and in this sense Mary learns* nothing exactly because of the reasons Lewis and Nemirow present (her new ability is non-verbal). But this argument is implausible partly because of the unusual notion of knowledge*. Furthermore, Jackson would have one more move in this game (had he not defined his 'knowledge' as "whatever can be told"). He could argue that, although it is not knowledge* that Mary acquires on seeing her first red object, what she gets is a basis for such knowledge*. Lewis says this in his formulation of ability hypothesis (Lewis 90); unfortunately Jackson does not join him on this.
I conclude that the knowledge argument can be ammended along the lines suggested by Lewis as a "basis for knowledge argument". In this form it may be defensible.
The Second Criticism
The criticism questioning (1) has been adopted by Harman (90): Mary did not know everything physical that there is to know about other people since she did not know "how the concept R [red] functions with respect to the perception of things that are red". (9) Harmans objection presumes that we should give a functionalist account of the content of concepts as well as of mental states.
Harman contrasts this kind of functionalism with Lewis functionalism which does not include a functionalist account of concepts. He also questions the ability of the latter to answer Jacksons criticism "since it is clearly false to say that the person blind from birth does not lack any information". (10) Consequently, Harman (90) approves Jackson's argument against traditional functionalism but constructs a broader 'functionalism of concepts' supposed to sustain Jackson's argument. (11) Truths about other people which escape third-person description do not necessarily "escape the physical story" if by 'physical story' one understands physicalism (and not a story about physicalism). Harman's argument shows that Jackson's argument does not reach physicalism of concepts.
The Turing Test for Perceptive Robots
The fact that Jacksons criticism does not succeed with physicalists of the second kind can be shown if we consider an imitation game similar to Turings classical test (12) but differing from it in one important respect. As in the tests original version, we try to find out which of the participants is a human and which an artificial intelligence. The difference is that the participants in the original Turing Test are computers judged for their ability to engage in a conversation whereas our game-partners are judged solely on the basis of their ability to discriminate qualia. We may call this test the Turing Test for Perceptive Robots.
In Turings article there is a remark about the possibility of constructing a machine able "to enjoy strawberries and cream". He refers to an attempt to construct one as "idiotic" but far from an impossibility. I am not particular about strawberries and cream, but there is nothing idiotic in constructing thousands of different machines detecting waves within visible and audible spectra or chemical substances one might taste, i.e. detecting qualia, which are used for different purposes including in the military and space programs. (These machines usually also detect impulses outside our possible sensory experience, but this is not our concern.)
Let me define a game some of these robots could play, called "The Black and White Mary Game". Some robots are trained exactly the way Jacksons Mary has been:
A). Although they are able in principle to detect visual inputs of the kind Mary would call red, they havent been actually exposed to this sort of experience in the past (like Mary hasnt seen anything red).
B). They have also extensive information about colors formulated in terms of physics (the way Mary has) i.e. in a third-person language.
At some point a red ripe tomato is demonstrated to the robot V (the way it was shown to Mary in Jackson's example).
What happens next?
Mary and robot V behave in exactly the same manner. Upon receiving the impulses (which, in Mary's case, we interpret as qualia) the robot "learns something about the world and visual experience of it" to repeat Jacksons sentence about Marys experience. (13) There is no way to sort out which reactions are produced by a robot (V) and which by a human (Mary). (14) This is a fatal blow to the Knowledge Argument.
Any Further Results?
The Turing Test for Perceptive Robots does not prove the reductionist conclusion that Mary and computer V do not differ. This conclusion would imply that there are only two choices:
Although Mary's and computer's behavior is not distinguishable by an external observer, the results say nothing about qualia understood in first-person Nagelian terms.
The intuitive appeal of Marys case that the robots case lacks does not come from any form of the Knowledge Argument. It originates from an aspect of the Black and White Marys case not covered by this argument, from the 'what it is like' problem. Computers do not have the first-person experience of qualia. Mary leaving her bedroom is supposed to become acquainted with them in a way specific to conscious subjects but this fact does not manifest itself functionally in any special way. The crux of Jackson's thesis is that they do, but the robot example shows in what way he is mistaken.
Schoemaker seems right arguing (in reference to the inverted qualia argument) that it gives us good reasons not to adopt functionalist criteria of verification. (15) The same is true about qualia in general. Functional differences are not enough to account for differences with regard to qualia.
To conclude this section, the Turings Test for Robots proves that the information argument does not determine solutions about qualia in any way. There is no peculiarity of Marys behaviour that a robot could not exhibit: the information Mary did not have is exactly the information robot V also lacked. The test shows that the Knowledge Argument against physicalism falls to pieces unless Jackson chooses to endorse an unattractive thesis that robots' behavior also disproves physicalism (that robots have an irreducible mental insight).
II. A Second Try
Now I shall take as given the following facts: Mary's and the computer's cognitive position is the same and their success in solving the correlation problem depends on their information-processing abilities, not on any abilities specific to humans that robots would a priori lack. The intuitiveness of the Knowledge Argument comes from some introspective, first-person experience that we expect Mary, but not a robot, to have.
In his "Epiphenomenal Qualia" Jackson emphasizes differences between his 'knowledge argument' and Nagel's question 'what it is like' in the following way: "I was not complaining that we weren't finding out what it is like to be Fred. I was complaining that there is something about his experience, a property of it, of which we were left ignorant". (16)
But Jackson is right only in a sense in which 'physical information' is defined as whatever can be told or given as a third person knowledge (17) we have rejected this claim in a previous section. Computer V, while playing "The Black and White Mary Game", acquires nothing else than the "physical information" Jackson is refering to. So, we are left ignorant about two different things concerning Mary's experience. One of them is information common to Fred, Mary and to the robot V; we have no reason not to call this sort of information physical. There is also a second aspect of Mary's experience, the one Nagel focuses on information of the kind "what it is like". I don't expect a computer to have this strictly first-person experience, and only this experience constitutes what seems 'not physical' in Mary's case.
Nagel makes it clear that it is impossible to analyze the first-person experience "in terms of any explanatory system of functional states, or intentional states, since those could be ascribed to robots or automata that behaved like people though they experienced nothing." (18) The first person statements are based on an immediate experience (or on a hermeneutics) of qualia, and not on proof of any kind.
This conclusion leaves us with all the numerous and well discussed problems that Nagel's 'what is it like' thesis produced. But Nagel's position has evolved substantially since 1974 when his "What is it like to be a bat" was first published. In his later and subtler approach Nagel tries to "juxtapose the internal and external or subjective and objective views" (19) emphasizing that neither of these perspectives is clearly superior. In particular the first person perspective cannot be reduced to objective, third-person knowledge since there are things which "can only be understood from the inside". (20) Points of view, characterised as "irreducible features of reality", provide the first-person perspective. This perspective cannot be stated in terms of external truth conditions. (21)
Jackson criticizes Nagel's position as "weak from a polemical point of view". (22) But since his alternative seems to fail I doubt if any polemically stronger argument in favor of strictly subjective qualities can be developed. Unlike Jackson, Nagel does not have illusions that there is a way of providing the evidence of first-person experience on the basis of third-person abilities, dispositions or behavior.
For Those Who Remain Faithful
In the remaining part of this paper I shall limit myself to following Terry Horgan's approach and to: "direct the subsequent discussion at those who consider it obvious, as I do, that the qualitative aspects of mentality are not definable functionally." (23)
Belief the qualitative aspects of mentality are not definable functionally lies behind Harman's functionalism of concepts which accepts qualia as complementary to the functional aspects. Harman adopts Dilthey's notion (he calls it Das Verstehen) which refers to the understanding non-reducible to the method of physical science in studying the mind. Das Verstehen refers to the irreducibly first-person experience necessary for understanding certain claims e.g. "Pain!" As Harman put it, obviously following Nagel: "You can know everything objective there is to know about a person without knowing what it is like to be that person". (24)
According to Harman's the language has its objective (or intersubjective) and subjective aspects. We are interested not only in transmitting objective, externally verifiable information, but we also care about other people's first-person experiences (it includes Das Verstehen). Hence, there are two different kinds of stories you can tell about an event like pain:
Harman's radical claim is that one may understand the objective story without understanding what it is like to be in pain, although these two different stories may satisfy the same criteria of confirmation. Lacking acquaintance with qualia one fails to understand something, which is different from not being able to confirm something. (25) A full understanding of the objective truth conditions of the word "red" does not give a full understanding of this word since a part of such understanding is knowing what red things look like. Mary's lack of subjective understanding how red things look means that she does not have a full understanding of something objective, namely, the proposition that something is red.
Harman targets an important misconception about our language as lacking interest in subjectivity. His criticism of positivist reductionism is based on an appreciation of two different facts: the physical facts about experience (e.g. physiology of seeing red) and an understanding of what it is like to see something red. (26) This leads him to a complementary structure of argument which require both: scientific third-person knowledge and Das Verstehen.
The issue of complementarity of our language is an aspect of a broader complementarity between the subjective and objective viewpoints. The issue, introduced to the contmeporary American philosophy by Nagel, has a longer history. Diltay's notion comes from Kant's First Critique. It was further developed in Fichte's introductions to philosophy of knowledge. Trough Husserl the issue became crucial to phenomenology and through Marx it reached Marxism as dialectics of subjectivity and objectivity. (27) In this paper I can not present these arguments I discuss these issues elsewhere (Boltuc 87, 88, 90).
My point is only to show that qualia present a problem for American philosophy: they are hard to defend whereas qualia reductionism is intuitively implausible. Kantian tradition, which might have been overemphasized in recent American ethical theory, remains neglected in metaphysics while it, as well as its later European developments, may be able to provided further grounds for complementarity of the subjective (28) and objective view-points.
One final remark, complementarity of the subjective and objective epistemic starting points has been discussed in Anglo-American philosophy, but these works have not been influential. It was presented in the clearest form by Gordon Globus, who later retracted these views for what seem to have been insufficient reasons. It was also discussed by Albert Shalom, probably thanks to his leanings towards phenomenological tradition. Finally, C. O. Evans in Britain endorsed something like complementarity while defending epistemic non-reducibility of consciousness. Unfortunately none of these arguments has been presented resolutely enough. Even Nagel's (79) persuasive argument on the subjective and objective perspectives stops short of endorsing an idea that epistemological complementarity is here to stay. This is why Harman's article, although it endorses complementarity only in a limited domain, is an important breakthrough in analytical philosophy.
|Notes(1) "Mary is confined to a black-and-white room, is
educated through black-and-white books and through lectures relayed on black-and-white
televison. In this way she learns everything there is to know about the physical nature of
the world. She knows all the physical facts about us and our environment, in a wide sense
of 'physical' which includes everything in completed physics, chemstry, and
neurophysiology, and all there is to know about the causal and relational facts cnsequent
upon all this, including of course functional roles. If physicalism is true, she knows all
there is to know. (...) It seems, however, that Mary does not know all there is to know.
For when she is let out of the black-and-white room or given a color television, she will
learn what it is like to see something red, say. This is rightly described as learning
she will not say 'ho, hum.' Hence physicalism is false." Jackson(86) p.291
(2) Jackson 86 p. 292
(4) As he put it: "Nothing you can tell can capture the
smell of a rose"Jackson 1982 p. 127
(5) Lewis (90) pp. 507-8. He relativizes this ststenent to:
"a certain class of worlds, which includes our actual world".
(6) I do not discuss Churchlands(85) famous criticisms in
any greater detail. The main points are answered in Jackson(86).
(7) Jackson 86 p. 293.
(8) Nemirow, Laurence : "Physicalism and the Cognitive Role
of Acquaintance" in Lycan 90, especially pp. 493-494
(9) Harman, Gilbert: The intrinsic Quality of
Experience", in: Philosophical Perspectives, 4 Action Theory and Philosophy of Mind,
(10) op. cit. p. 46
(11) Later Harman (91/93) attempts to solve the problem of qualia
outside of the traditional functionalist framework (through the notion of meaning seen as
not fully chartacterised by objective truth conditions).
(12) Turing, A.M.: "Computing Machinery and
Intelligence" Mind, vol. LIX No 236 (1950)
(13) Jackson, Frank 82 p. 130
(14) There is a minor inconvenience in the above formulation of
the Turing Test for Robots. If the robot had been constructed according to the
black-and-white-Mary-scheme, it would not be able to classify the inputs it had
received. We may call this lack of knowledge a correlation problem. So,
before the robot says "I enjoy seeing red" it needs to know how the new impulse
that it registers is called (namely "red", rather than e.g. "green").
Robot V, as well as Mary, need some more information to establish that their experience is
actually the experience called "red", though in the most cases there is a way a
robot, as well as a human, might discover this new piece of information, for example if
they can recognize ripe tomatoes (e.g. by shape) and if it knows that most ripe tomatoes
are red. But only after some information-processing, having eliminated the hypothesis that
the tomato is green, if there are good reasons to believe it is ripe independent of a new
quale, they may establish an inductive hypothesis that THIS is how red things
look. As David Lewis pointed out "One might well know what an experience is like
under one description, but not under another." This looks like a version of Frege's
famous Morning Star/Evening Star dilemma. In the next sentence Lewis opens an issue of
knowledge under no description whatsoever, which I do not discuss. "One might even
know what some experience is like, but not under any description whatsoever unless
it be some rather trivial description like 'that queer taste that I am imagining right
now'". see in: Lewis (90) p.515 The correlation problem is not a part of the
Knowledge Argument (Lewis presents it as one of the "ways to miss the point" of
the Knowledge Argument).
(15) Schoemaker, Sydney: "The inverted spectra" p.
(16) Jackson 82 p.132
(17) As he put it: "Nothing you could tell of a
physical sort captures the smell of a rose." (underlining mine)Jackson 82 p. 127. But
the physical information can hardly be defined just as '"what you can tell".
(18) Nagel 74 p. 436
(19) Nagel 86p.4
(20) op.cit. p. 18
(21) op.cit. p.30 This shows that Lewis' refutation of the
knowledge argument in terms of truth conditions (Lewis 90 p. ) does not necessarily refute
(22) Jackson 82 loc. cit.
(23) p. 459
(24) Harman, Gilbert: op. cit slides, topic 3. I refer directly
to the lecture though the similar content can be found in his article.
(25) Hence, the first one is not knowledge by acquaintance.
(26) A person blind from birth does not understand this
second aspect because she "does not have full enough concept of what it is like for
something to be red" op. cit. sec. 27.
(27) See: Lectorsky: Abdildin: Siemek:
(28) Nagel's argument that first-preson viewpoint is objective is
not forgortten; yet I follow the general usage of these terms.
Boltuc, Piotr (1987): "Introduction to Complementary Philosophy" Colloquia Communia (in Polish) 3-4 pp. 259-64.
Boltuc, Piotr (1988): "In the Beginning was a Contradiction: Problems of Philosophy of Subject and Object." Dialectics and Humanism 3-4/1988 pp. 177-186.
Boltuc, Piotr (1988): "Person as a Locus of Premanence: Towards Albert Shalom's Metaphysics" Dialectics and Humanism 2/1990 pp. 213-234.
Churchland, Paul (1985): "Reduction, Qualia, and the Direct Interpretation of Brain States" The Journal of Philosophy pp. 8-28.
Evans, C.O. (1970): The Subject of Consciousness London, Murrihead
Globus, Gordon, Maxwell, Grower, Savodnik, Irvine (1976): Consciousness and the Brain -- A Scientific and Philosophical Inquiry NY.
Harman, Gilbert (1990): "The Intrinsic Quality of Experience" in: "Philosophical Perspectives 4, Action Theory and Philosophy of Mind" pp. 31-52.
Harman Gilbert (1991/3): "Can Science Understand the Mind? in: Harman, Gilbert (ed.): "Conceptions of the Mind: Essays in Honor of George Miller", Erlbaum Publishing House, New York1993. (also transparencies for his lecture "Can Science Understand the Mind?" Princeton University, 3 October 1991).
Horgan, Terence (1984): "Functionalism, Qualia and the Inverted Scectrum" Philosophy and Phenomenological Research XLIV/4.
Jackson, Frank (1982): "Epiphenomenal Qualia" Philosophical Quarterly XXXII
Jackson, Frank (1986): "What Mary Didn't Know" Journal of Philosophy LXXXIII
Lektorsky, Nikolaj: "Subject, Object, Knowledge", Nauka, Moscow 1980
Lewis, David (1983): "Mad Pain and Martian Pain," with the "Postscript". "Philosophical Papers" vol.1 Oxford Univ. Press.
Lewis David (1990): "What Experience Teaches" in: Lycan William: "Mind and Cognition" Cambridge MA,1990 pp. 499-519.
Margolis, Joseph (1978): "Persons and Minds. The Perspectives of Nonreductive Materialism" Boston.
Nagel, Thomas (1974): "What is it like to be a Bat" Philosophical Review LXXXIII pp.435-450.
Nagel, Thomas (1979): "Subjective and Objective" in his: Mortal Questions Cambridge University.
Nagel, Thomas (1986): "The View from Nowhere", Oxford Univ. Press.
Nemirow, Laurence (1990): "Physicalism and the Cognitive Role of Acquaintance" in: Lycan William: "Mind and Cognition" Cambridge MA, 1990.
Shalom, Albert (1985): The Body/Mind Framework and The Problem of Personal Identity Humanities Press International, Atlantic Highlands.
Shoemaker, Sidney (1982): "The Inverted Spectrum" The Journal of Philosophy LXXIX/7 pp. 357-381.
Turing, A. M. (1950): "Computing Machinery and Intelligence" Mind LIX/236.